Aiming to explore the effect of fractional-order derivatives (FOD) combined with support vector machine discriminant analysis-random forest model (SVMDA-RF) on hyperspectral monitoring of desert soil organic matter content (SOM). The desert soil samples collected in the Sde Boker area of Israel were analyzed. These soil samples were through pretreatment, physical and chemical analysis, soil classification (divided into sandy soil (SS) and clay loam soil (CLS)), indoor spectral acquisition and spectral resampling (interval 10nm). In order to avoid the influence of soil quality on the inversion model, the support vector machines discriminant analysis (SVMAD) was established based on the average reflectance of the spectrum. The spectral reflectance was processed by 0~2 order (interval 0.2) FOD. Then NDI was constructed by using the spectral data that through fractional order derivatives processing and the twodimensional correlation between SOM and NDI was analyzed. In order to obtain all different FOD enhancedNDI, the highest coefficient of determination (R2) of 0-order NDI was used as the threshold (sand soil R2>0.901, clay loam soil R2>0.763). By using the different FOD enhanced-NDI to establish random forest (RF) models. All models based on different soils were compared and analyzed, the best models of different soils were combined to establish the SVMDA-RF model. The results showed that SVMDA based on spectral average reflectance, the classification rate of soil texture could reach 100%. Fractional order derivatives coupling normalized spectral index, which had ability to amplify SOMrelated implicit information between bands, but it had different effects on different soils. For the paper, clay loam soil was superior to sandy soil, and two soils of the FOD-enhanced sensitive index peaked at 0.6-order, but the number of sensitive index of clay loam soil was much larger than that of sandy soil. In the sandy soil RF models, the model based on 1.2-order NDI was the best（R2C=0.962，R2P=0.920，RMSEP was 0.435g/kg，and RPD was 3.658）. In the loam RF models, the model based on 0.6-order NDI was the best（R2C=0.942，R2P=0.944，RMSEP was 0.554g/kg，and RPD was 4.316）. Combining the optimal models of the two soils to get the high-precision SVMDA-RF model, R2P=0.979，RMSEP was 0.481g/kg，and RPD was 7.004. The model could provide effective support for quickly assessing the desert soil types and fertility.